| New file |
| | |
| | | # network architecture |
| | | frontend: default |
| | | frontend_conf: |
| | | n_fft: 400 |
| | | win_length: 400 |
| | | hop_length: 160 |
| | | |
| | | # encoder related |
| | | asr_encoder: conformer |
| | | asr_encoder_conf: |
| | | output_size: 256 # dimension of attention |
| | | attention_heads: 4 |
| | | linear_units: 2048 # the number of units of position-wise feed forward |
| | | num_blocks: 12 # the number of encoder blocks |
| | | dropout_rate: 0.1 |
| | | positional_dropout_rate: 0.1 |
| | | attention_dropout_rate: 0.0 |
| | | input_layer: conv2d # encoder architecture type |
| | | normalize_before: true |
| | | pos_enc_layer_type: rel_pos |
| | | selfattention_layer_type: rel_selfattn |
| | | activation_type: swish |
| | | macaron_style: true |
| | | use_cnn_module: true |
| | | cnn_module_kernel: 15 |
| | | |
| | | spk_encoder: resnet34_diar |
| | | spk_encoder_conf: |
| | | use_head_conv: true |
| | | batchnorm_momentum: 0.5 |
| | | use_head_maxpool: false |
| | | num_nodes_pooling_layer: 256 |
| | | layers_in_block: |
| | | - 3 |
| | | - 4 |
| | | - 6 |
| | | - 3 |
| | | filters_in_block: |
| | | - 32 |
| | | - 64 |
| | | - 128 |
| | | - 256 |
| | | pooling_type: statistic |
| | | num_nodes_resnet1: 256 |
| | | num_nodes_last_layer: 256 |
| | | batchnorm_momentum: 0.5 |
| | | |
| | | # decoder related |
| | | decoder: sa_decoder |
| | | decoder_conf: |
| | | attention_heads: 4 |
| | | linear_units: 2048 |
| | | asr_num_blocks: 6 |
| | | spk_num_blocks: 3 |
| | | dropout_rate: 0.1 |
| | | positional_dropout_rate: 0.1 |
| | | self_attention_dropout_rate: 0.0 |
| | | src_attention_dropout_rate: 0.0 |
| | | |
| | | # hybrid CTC/attention |
| | | model_conf: |
| | | spk_weight: 0.5 |
| | | ctc_weight: 0.3 |
| | | lsm_weight: 0.1 # label smoothing option |
| | | length_normalized_loss: false |
| | | |
| | | ctc_conf: |
| | | ignore_nan_grad: true |
| | | |
| | | # minibatch related |
| | | batch_type: numel |
| | | batch_bins: 10000000 |
| | | |
| | | # optimization related |
| | | accum_grad: 1 |
| | | grad_clip: 5 |
| | | max_epoch: 60 |
| | | val_scheduler_criterion: |
| | | - valid |
| | | - loss |
| | | best_model_criterion: |
| | | - - valid |
| | | - acc |
| | | - max |
| | | - - valid |
| | | - acc_spk |
| | | - max |
| | | - - valid |
| | | - loss |
| | | - min |
| | | keep_nbest_models: 10 |
| | | |
| | | optim: adam |
| | | optim_conf: |
| | | lr: 0.0005 |
| | | scheduler: warmuplr |
| | | scheduler_conf: |
| | | warmup_steps: 8000 |
| | | |
| | | specaug: specaug |
| | | specaug_conf: |
| | | apply_time_warp: true |
| | | time_warp_window: 5 |
| | | time_warp_mode: bicubic |
| | | apply_freq_mask: true |
| | | freq_mask_width_range: |
| | | - 0 |
| | | - 30 |
| | | num_freq_mask: 2 |
| | | apply_time_mask: true |
| | | time_mask_width_range: |
| | | - 0 |
| | | - 40 |
| | | num_time_mask: 2 |
| | | |